The use of intention to treat cohorts is the standard of nearly every study.
If Missling does not want to be laughed at on Monday he will need to study the difference between ITT and PP (per protocol). Per protocol analysis is useful --- how did patients who completed the study perform. However, the FDA and EMA do not consider PP as a valid population for most studies. > 10% of the treated patients dropped out, some early and some after, say, 9 months. The gold standard hurdle that everyone including Biogen and Anavex must clear is significance using ITT and accounting for dropouts by various imputation methods. If you ever read the statistical analysis from the FDA, often 100 pages, it is fascinating the detail used to evaluate this. They expect the company to do imputation sensitivity and show their data with various imputation techniques. Patients who drop out are treatment failures, especially if due to a side effect --- their presence can't be ignored.
Lets throw out numbers: n=170 placebo and n=338 A273 --- this is the ITT group. They might validly claim n=168 and n=335 as 5 patients did not take any medication (they signed the ICF, passed the screen and then withdrew at or before the visit where they would have received the medication). Either of these groups would be valid as ITT (the second being a common modified ITT usually acceptable if numbers are small as they are). The slide 21 data is apparently from the PP cohort n=161 and n=301 of protocol completers. They are part of a pre-specified analysis but not the primary analysis. How can we impute data if a patient drops out??? The two most used methods are LOCF (last observation carried forward) and MMRM (mixed effect model repeated measures). LOCF used to be the most used -- its simple and you just use the last numbers (i.e baseline or 36 weeks or whatever) and assume it is the last number. It is still sometimes used in pain studies but not used for progressive disorders because it rewards the treated group for adverse event dropouts. MMRM is a fancy phrase but basically means data is imputed by assuming the patient left the cohort by random and if they stayed in the cohort they would have done about the same oover the next time point as others. So a patient who lest after 12 weeks is modeled at 48 weeks by assuming they progressed or improved at the same rate as the rest of their group (sometimes adjusted by demographics -- all spelled out in the SAP). So the FDA/EMA/TGA will expect to see ITT with MMRM imputation. Anything else would have been fine for a Phase 2b but not a phase 3. Missling knows this. I hope he does not try to be cute and try to hype PP data if the ITT data is not first presented. An ITT miss with PP hit is a phase 3 fail but not a 2b total loss.
Responder/non-responder data can also be imputed they just have different techniques compared to parametric mean / std deviation as the ADAs-Cog and ADCS-ADL actual scores and change would be.
So on Monday I want to hear: Using the ITT group what was the baseline mean ADAS-cog for the placebo and total A273 group (or 30/50 in a minute) using MMRM or other clearly stated imputation technique and what was the 48 week mean ADAS-Cog for the two groups. N should be 168 (or 170) for placebo and 335 (338) for treated. The statistical test should be t-test. Ditto for ADCS-ADL. The p value, with co-primary endpoints, should be <0.025 for both.
What if they have time to present endpoint data for 30 and 50 mg? That would be great. Patients are defined by entry dose so a 50 mg who went down to 30 mg counts as a 50 mg (I don't write the rules). Definitions of ITT groups and PP groups are the same as above and imputation would be the same. So each group would have about 168 patients for ITT. He should clearly present n for the ITT group and n at 48 weeks (especially if wanting to tout the PP results). Use the ANOVA test rather than T-test as is standard for 3 or more groups running simultaneously. If he uses the Jonkheere Trend test rather than ANOVA (as in the episodic memory PDD result) then realize that Missling is confirming that this was a P2b and not a real P3. His audience became this message board and not the regulatory agencies.
The above is what I want to hear from Missling. If he prefers to talk about the responder analysis and skips the mean change then we can assume the study failed in regards as a pivotal study though much useful information will be obtained to design the next 1-2 phase 3 studies. Maybe he could justify why 0.5 points was chosen for the ADAS-Cog besides that that threshold led to a good response while 3 points did not.
Failure to do appropriate stats will be easily caught. If I could say anything to Missling it would be "Please present the data appropriately and if we did not achieve enough success lay out a plan and don't overhype" Missling could very easily pull one over on some and claim success to message board poster applause. AF (or his people) and JB will see through poorly presented stats and be merciless in their criticism. A phase 3 is carefully evaluated by experienced statisticians at the FDA/EMA/TGA who can parse through the BS to find the real result and they will present to the others at the FDA who make decisions and not be ignored. .
